Why not paste every PDF?

Week 6 · RAG and evals · about 3 minutes

What you will be able to do

Stop stuffing Helix's whole policy library into the prompt — bigger context costs more and answers worse.

The idea

A bigger context is not a better brain. Extra pages raise cost, latency, and the chance the model attends to the wrong paragraph.

Retrieve a few relevant chunks. Paste only what the question needs.

# Bad: prompt = "\n".join(all_pdfs) + question
# Good:
chunks = retrieve(question, k=5)
prompt = format_context(chunks) + question

If the answer needs a library, use retrieval — not a paste.

Key takeaways

• Whole-library prompts cost more and answer worse.
• Index once; retrieve per question.
• Small relevant context beats huge noise.

Part of the Forward Deployed Engineer roadmap. New to the role? Start with what a Forward Deployed Engineer does.